SIDERUNE

Forensic accountability infrastructure

Your AI agent took action. Can you prove it was authorized?

Reconstruct the complete chain from human authority to real-world effect.

See how it works contact@siderune.io

Canonical example

A $482 refund against a $100 authority limit

A human authorizes Agent A to issue refunds up to $100. Agent A delegates to Agent B. Agent B requests a $482 refund. The external system records a $482 effect. This is what SIDERUNE reconstructs.

  1. Human grants authority
  2. Authority ≤ $100
  3. Agent A holds authority
  4. Delegated to Agent B inherits ≤ $100
  5. Requested $482
  6. Actual effect $482
  7. EXCEEDED violation
  8. Evidence Verified hash chain intact

Authorized limit $100. Requested $482. Actual effect $482, $382 beyond the limit. Violation: true. Root cause: the delegated agent exceeded its inherited authority — reconstructed and verified against the evidence chain.

What it answers

Seven questions, answered from evidence

  1. Who authorized the action?
  2. What authority did the agent actually have?
  3. How was that authority delegated?
  4. What action was requested?
  5. What actually happened in the external system?
  6. Was the real effect within the authority that existed at that moment?
  7. Can the reconstruction and evidence be verified?

How it works

From event to verifiable evidence

SIDERUNE captures events as they occur, evaluates them against the authority that was actually granted and delegated, compares requested action to actual effect, and reconstructs the full chain from tamper-evident evidence.

  1. Event capture
  2. Authority & delegation
  3. Actual effect
  4. Reconstruction
  5. Verifiable evidence

SIDERUNE reconstructs and verifies the chain from events your systems and agents already produce. It does not currently offer a managed policy-enforcement gateway, and reconstruction is only as complete as the events supplied to it.

Built for

Who relies on this

Teams responsible for what an AI agent was allowed to do, and for proving what it actually did.

  • CTOs
  • Engineering leaders
  • Security leaders
  • AI infrastructure teams

Evidence integrity

Built on verifiable evidence, not assertions

  • SHA-256 hash chaining

    Each recorded event is linked to the one before it, so the sequence itself can be checked for gaps or tampering.

  • Ed25519 signatures

    Evidence is signed, so its origin and integrity can be independently verified rather than taken on trust.

  • Tamper-evident evidence

    Historical events are designed so that any after-the-fact alteration is detectable, not merely discouraged.

  • Authority & delegation reconstruction

    The chain from human authority through every delegation step to the acting agent is reconstructed explicitly.

  • Actual-effect comparison

    Requested action, actual execution, and actual real-world effect are tracked as distinct facts and compared against the authority that existed at the time.

SIDERUNE is an internal accountability and forensic reconstruction layer. These are engineering properties of the system, not a claim of external certification, regulatory compliance, or guarantee of outcome.

Get in touch

Talk to us about your agent accountability chain

We're working directly with early teams building consequential AI agent systems.

contact@siderune.io